Have an idea?

Visit Sawtooth Software Feedback to share your ideas on how we can improve our products.

Analyzing a CBC in which one attribute is a constructed list


It was mentioned here: https://sawtoothsoftware.com/forum/7553/constructed-used-populate-levels-attribute-different-brands that it is is possible to pass a constructed list for the levels of an attribute for a CBC exercise (such as brands), but that the analysis on the back end is tricky.

Are there any materials or further information that could be provided to guide me in the analytical process? Roughly, is it a matter of running a separate HB among respondents who were shown the same sub-list of brands?

I'm using SSI Web v8.4.8, for what it's worth.

EDIT: I found this answer, which helps quite a bit: https://sawtoothsoftware.com/forum/13824/choose-efficient-tasks-with-constructed-list-one-attribute
asked Feb 13, 2018 by Phil (145 points)
edited Feb 13, 2018 by Phil

1 Answer

0 votes
I cannot think of a handy document that describes how to do constructed lists for brands in CBC.  

Certainly ACBC (Adaptive CBC) has capabilities built into the software for doing this and with supporting documentation.

With CBC and constructed lists for brands (a power trick), the big gotcha is that the list of brands has to be identical length for each respondent.  (That's because CBC uses an experimental design that you generate prior to fielding the study and that is uploaded as a database of designs to the web server.)

There are multiple approaches to analysis.  And, the approach you take depends on whether the brands you selected for a respondent were due to preference or availability within the respondent's region.

You typically do NOT have to break out the respondents into subsets of respondents who received the same brands; there typically are ways to reorganize the data for analysis such that all respondents are kept together.  But, in the case of just a few regions (say 4 or fewer) dictating which brands are available within the region and in the case that each region has around 500 or more respondents, it would be appropriate and easy to break out the respondents by region into different data sets and to run analysis separately within each of those groupings of respondents.
answered Feb 13, 2018 by Bryan Orme Platinum Sawtooth Software, Inc. (173,090 points)
Thanks Bryan. Given you answer on the question I linked at the end, I'm leaning towards pushing back against the usage of a constructed list, or that we move this into an ACBC exercise.
ACBC tends to work well if the total number of attributes is 4 or greater.
There are 6 attributes in total, so this wouldn't be a problem.